Question

I would like to log dynamic tag values for the DistrbutionSummary metric, how can I do that?

在以下例子中,有一个称为“我方”的标签,我希望根据所收到的数据,将标记的价值记录在传单上。我可能想把分发摘要记录下来的多个方面,除非实时,否则我不会知道这些标签价值。

DistributionSummary histogram = DistributionSummary.builder("my_histogram")
            .description("Example histogram")
            .baseUnit("ms")
            .publishPercentiles(0.5, 0.95)
            .tags("myTag", "myTagVal")
            .serviceLevelObjectives(100.0, 500.0, 1000.0)
            .distributionStatisticExpiry(Duration.ofMinutes(5))
            .distributionStatisticBufferLength(3)
            .register(meterRegistry);

For eg., Metrics.counter from micrometer has a way to pass the value for the tags on the fly while incrementing or recording the value. how can we do the same in DistributionSummary?

Metrics.counter("eventsProcessed", "myTag", anyValue).increment(count);

Answer 1

你在你的问题中采取同样的方式:

DistributionSummary.builder("test.distribution")
        .tag("static", "abc")
        .tag("dynamic", tagValue) // comes from a variable dynamically
        .register(meterRegistry)
        .record(value);

<编码>register Basic means create NewOrGetExmain。而且,自微观参数1.12.0以来,我们有。为此:

MeterProvider<DistributionSummary> distributionProvider = DistributionSummary.builder("test.distribution")
        .tag("static", "abc")
        .withRegistry(registry);

// [...]

distributionProvider.withTags(Tags.of("dynamic", tagValue)).record(value);

其他几个想法:

What you did with the Counter is also possible with the DistributionSummary: registry.summary("test", "dynamic", tagValue).record(value);

Since seemingly you are measuring latency, I would highly recommend using a Timer instead, here s why
As others also advised before, please double-check if you don;t have high-cardinality tags (unnormalized user input included), here s why

Answer 2

我曾遇到过这一问题,并通过阅读源码找到解决办法。你们可以采用以下方法实现动态调整。 The pseudocode:

private final String LISTED_MARKET_SSE = "SSE";
private final String LISTED_MARKET_SZE = "SZE";

@Override
public void bindTo(@NonNull MeterRegistry meterRegistry) {
    String tagListedMarket = "listed_market";
    String upstream7101LatencyName = "kafkaConsumer.upstream.7101.latency";
    upstream7101LatencySummary = DistributionSummary.builder(upstream7101LatencyName)
            .baseUnit("ms")
            .tag(tagListedMarket, "")
            .publishPercentiles(0.99, 0.95, 0.5)
            .register(meterRegistry);
    upstream7101LatencySummarySZE = DistributionSummary.builder(upstream7101LatencyName)
            .baseUnit("ms")
            .tag(tagListedMarket, LISTED_MARKET_SZE)
            .publishPercentiles(0.99, 0.95, 0.5)
            .register(meterRegistry);
    upstream7101LatencySummarySSE = DistributionSummary.builder(upstream7101LatencyName)
            .baseUnit("ms")
            .tag(tagListedMarket, LISTED_MARKET_SSE)
            .publishPercentiles(0.99, 0.95, 0.5)
            .register(meterRegistry);
}

然后使用:

    if (LISTED_MARKET_SSE.equals(listedMarket)) {
        upstream7101LatencySummarySSE.record(amount);
    }
    else if (LISTED_MARKET_SZE.equals(listedMarket)) {
        upstream7101LatencySummarySZE.record(amount);
    }
    else {
        upstream7101LatencySummary.record(amount);
    }

友情链接